Every artificial intelligence (AI) conversation in South African business right now starts in the same place: what can it do? Vendors demo capabilities, conferences rank models and boards ask for “an AI strategy” as if the technology were a substance you could order by the litre.
After two years immersed in how large African financial institutions actually adopt AI, I want to argue that the capability question is the wrong first question.
The right one comes from a 50-year-old idea ecological psychologists call an affordance, and it is quietly the most practical concept in responsible AI.
An affordance is not a feature. It is a possibility for action that exists in the relationship between a technology and a particular group of people, in a particular context, with particular skills and goals. A staircase affords climbing, but not to everyone, and not in every building.
The same generative AI service affords fraud-pattern triage to one bank team, first-draft drudgery-removal to another, and nothing at all to a third that lacks the data, the permission or the confidence to use it.
AI adoption is never uniform, because using these tools well is not a skill alone; it is skill compounded by attitude and habit of mind.
The action possibility lives in the relation, not in the product sheet. That is why two organisations can buy identical licences and get utterly different outcomes, and why “what can AI do?” has no useful answer until you ask, “for whom, under what conditions, here?”
This is where “AI for good” stops being a slogan and becomes a design discipline. If value lives in the relation, then the ethical centre of gravity shifts from the model to the people and conditions around it.
Where AI lands well inside large financial institutions, a pattern is unmistakable. Three human conditions are met: capable hands (tools deliberately distributed to people equipped to use them), honest measurement (task-level gains that are real but routinely more modest than the slogans; headline percentages rarely survive contact with the person who owns the numbers), and communities that convert curiosity into practice, growing by demonstration rather than mandate.
Leaders increasingly demand exactly that: show me, not tell me. And adoption is never uniform, because using these tools well is not a skill alone; it is skill compounded by attitude and habit of mind. Affordances put people (their skills, their confidence, their communities) inside the technology question, which is exactly where a for-good agenda needs them.
The affordance lens also exposes the harms that capability-talk hides. The most consequential governance risk in enterprise AI today is not a rogue chatbot; it is functionality that arrives switched on, enabled by vendors by default: AI capabilities activated inside routine platform upgrades before any gate, policy or ethics committee ever saw them.
In affordance terms: a possibility for action was inserted into thousands of working relationships without anyone deciding it should be. If your responsible-AI programme audits models but not defaults, it is auditing the brochure, not the building. The fix is unglamorous and effective: inventory default-enabled AI across the vendor estate, contract for disclosure and off-switches, and read upgrade notes as risk documents.
And for Africa, specifically, affordances explain something painful that access-talk obscures: why the AI wave sorts us. Pan-African institutions operating across many markets under one strategy find that the wave arrives early where years of cloud migration, data clean-up and regulatory groundwork have prepared the relation, and late where they have not.
Markets without hyper-scaler cloud presence lag not because their people are less capable but because the conditions that turn capability into affordance are missing.
“AI for good” on this continent, therefore, means building ground: data-protection regimes with enforcement teeth, cloud infrastructure, skills pipelines: the unphotogenic substrate that decides who actually gets to act on the possibilities everyone is selling.
So, here is the affordance test I would put to any leadership team spending money on AI this year. Not “what can it do?” but: who, specifically, will this afford new action to, and who will it not? What conditions (data, skills, permission, funding) must hold before the possibility is actualised? What did we measure, honestly, when we tried? And what has been switched on around our people without a decision?
Organisations that can answer those four questions are doing “AI for good” in the only sense that survives contact with reality: technology whose possibilities are opened deliberately, to people equipped to act on them, under conditions someone actually built. Everything else is capability worship; and capability, on its own, has never been good for anyone.
* Billy Mashele lectures at the University of Cape Town, where he researches technology adoption and organisational change in financial services. He writes here in his personal capacity.
Share
